Gender analysis from fisheye cameras

Hello! I work in the retail business. To improve our service to our customers, we decide to deploy cameras in the stores to detect the percentage of male and female shoppers we have, and what are their preferences for the merchandise.

We want to install some fisheye cameras on the ceiling, but we are not sure of the accuracy of the detection. I asked one consultant from Tyco, he told me that the accuracy is more than 99%. Is it possible? We doubt it. Because even humans would make mistakes on appearance, not to mention the camera from the top angle.

Do you guys have any thoughts? We are making the final evaluation to deploy the cameras for data collection, need your advice on it.

System integrator here. 99% is impossible. The accuracy depends on how the software has been well trained. I would say the best accuracy for telling gender from fisheye camera might be around 95 to 96% according to my experience. In the other hand, I think more than 90% should be quite enough, don't you think so?